Evaluation of weather data sources for estimating energy demand in industrial chillers

Accurate estimation of energy consumption in industrial cooling systems is essential for improving efficiency and reducing operational costs, particularly in data centers. This paper investigates the impact of different meteorological data sources on the prediction of energy demand in an industrial chiller. Specifically, we compare traditional groundbased weather measurements with data derived from multiple Numerical Weather Prediction (NWP) models. A linear regression model is developed to estimate daily energy consumption based on temperature, humidity, and wind speed, including lagged variables. The predicted values are then used within an anomaly detection framework to identify periods of excessive energy usage. Experimental results show that NWP-based inputs achieve performance comparable to ground measurements, with only marginal differences in prediction accuracy. Furthermore, the proposed approach enables the identification of abnormal consumption patterns, highlighting the potential of NWP data as a reliable and scalable alternative for energy monitoring applications.